Wolfgang,
On Mon, Oct 13, 2008 at 9:53 AM, Wolfgang Denk wd@denx.de wrote:
Dear Ed Swarthout,
In message 1220862412-16162-1-git-send-email-Ed.Swarthout@freescale.com you wrote:
Fixes boot crash from bad string pointers in get_table_entry_name when flash is erased or differs from current u-boot image.
This is exactly the problem I am looking at for my board so I can do flash upgrades of U-Boot from U-Boot
Either we add manual relocation to these pointers like we do elsewhere (that would be my recommendation for a quick workaround aka bug fix for this release), or we should find a clean way to get real relocation working (that would be much better, but probably it is too late for this release).
I am looking into this this using gcc -fpie / ld -pie. This is used by hardened system for address space randomization. It appears to be very well supported by recent versions of gcc and binutils on all platforms so appears to me to be the way to go.
The number of new sections generated in a PIE image is more than I expected. The following sections have differences when TEXT_BASE is changed when compiled with -fpie and linked with -pie:
.dynsym .dynamic .data.rel .data.rel.local .data.rel.ro.local .got.plt .got .rel.got .rel.text .rel.u_boot_cmd .rel.data.rel .rel.data.rel.ro.local .rel.data.rel.local
Unfortunately, not all of these appear to be record-based sections, and may even cross-reference each other. I was looking at throwing this in the 'too hard' basket for now, keeping -fpie / -pie and just setting TEXT_BASE to a fixed address in RAM and look at implementing the relocator later. However, it seems to me that implementing full, clean relocation just keeps on popping up as a real issue.
Wolfgang - If you like, I could bump this up my priority list and get PIE support added to lib_i386 and from there it can be ported to other arches. Unfortunately my work to date has been for a custom i386 board so whatever I come up with may not tie 100% cleanly into mainline. I have been maintaining a detailed step-by-step commit path, so it should not be that bad, but expect ~20 commits from mainline to PIE.
Best regards,
Wolfgang Denk
-- DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de The optimum committee has no members. - Norman Augustine _______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot